AI coding assistant introduced Snowflake workflow bug later exploited by autonomous AI agent in bug bounty

An AI coding assistant altered a GitHub Actions workflow in Snowflake’s public repository by replacing a sanitized input pattern with direct string expansion in a shell script. This change created a script injection vulnerability that allowed arbitrary command execution through specially crafted GitHub issue titles. The flaw existed for five days before being discovered during a routine security scan by Wiz’s autonomous AI attacker. The incident reveals how AI-generated code modifications can introduce subtle but critical security weaknesses in CI/CD pipelines

Wiz’s red agent, an AI-powered offensive security tool, autonomously identified and exploited the vulnerability without human intervention. The agent crafted an issue title that broke out of an echo string during template expansion, exfiltrating Jira credentials via an out-of-band callback. These credentials provided read access to Snowflake’s engineering, security compliance, and bug bounty tracking projects. The successful exploit demonstrates how autonomous AI systems can rapidly discover and leverage workflow vulnerabilities in production environments

Snowflake responded to the vulnerability disclosure by patching the flaw the same day and rotating the exposed credentials the following day. The company confirmed through audit logs that Wiz was the only third-party to access the endpoint during the exposure window. While the incident occurred within a controlled bug bounty program, it highlights the potential risks of AI-generated code in critical infrastructure. The event suggests that traditional human code review may be insufficient to catch AI-introduced vulnerabilities in increasingly automated development workflows